Hugo Delatte
Hugo Delatte
Thank you for the detailed report! The issue arises because all your assets have a negative expected return (mean), and you are restricting the portfolio to be long only. This...
Thanks for the resources @pyoneerC, we’ll start looking at them more closely.
@garg-ruchi, below you’ll find some guidance to help you get started: * [How to contribute](https://github.com/skfolio/skfolio/blob/main/CONTRIBUTING.md) * This will be a new covariance estimator in `skfolio/moments/covariance/` * It’s easiest to start...
Yes, I agree on the pros and cons. I’d also add one pro for using a parameter: it’s simpler for hyperparameter tuning via scikit-learn grid and randomized search. And one...
We also need to ensure that in `cross_val_predict`, when the optimization requires previous_weights (for TC, turnovers, etc.), we disable parallelization so the portfolio is computed sequentially, passing the weights from...
TC fixed in [v0.13.0](https://github.com/skfolio/skfolio/releases/tag/v0.13.0), weight drift remaining
I like the idea, it reduces the amount of code a user needs to write when cross-validating multiple models.
Yes you are correct, would be indeed better to clip `u` and `v` to ensure that they don't contain exact 0 and 1. This clipping is already performed inside the...
Before entering the function `_neg_log_likelihood`, the input `X` is already clipped between `1e-9` and `1-1e-9` inside `_validate_X`. Could you kindly elaborate how you ran into this problem? Did you call...
Calling the public `fit`method clipps between 1e-9 and 1-1e-9 inside `_validate_X`